Michael Fischer's 1994 recordings of TR-808 serial 103852 (individual outputs, LEVEL at
maximum, 16-bit 44.1 kHz; docs/RolandTR808_reference_samples/TR808.TXT) against
the generated cores run at 4x oversampling and decimated to 44.1 kHz, and against full-card LTspice on the
same netlist where a trace exists. The recording chain is unknown, so every trace is onset-aligned and
peak-normalised and the comparison is of envelope, decay and spectrum - not a null. Decay times are to
1/10 (t20) and 1/100 (t40) of the peak, the manual's own convention. env rms diff is the rms
difference of the two envelopes in dB over the part of the recording above -40 dB: 0 would be the same
envelope. A SECOND UNIT, serial 209265 (docs/RolandTR808_synthmania_209265), is drawn in
grey where its file has clean hits: a performance with the pot turned while it plays, so it carries no knob position
and no level, but it says whether a difference from 103852 is ours or is one machine against another.